Pseudo cycles insertion
Description
This transformation turns straight pieces of code into cycles which has
only 1 trip count. This introduces some artificial paths into CFG as
well.
Transformed Code
See Appendix A. Demo for original sources.
public _Process
.align 16
_Process proc near
push EBP
push EBX
push EDI
push ESI
sub ESP, 12
mov EAX, DWORD PTR [ESP + 44]
mov DWORD PTR [ESP + 4], EAX
test EAX, EAX
jle .LBB2_13
.LBB2_1: ; %entry.cBB_crit_edge
mov EAX, DWORD PTR [ESP + 40]
mov ECX, 4294967295
.align 16
.LBB2_2: ; %cBB
inc ECX
je .LBB2_14
.LBB2_3: ; %bb.loopexit
mov EAX, DWORD PTR [ESP + 32]
lea EAX, DWORD PTR [EAX + 80]
mov DWORD PTR [ESP], EAX
xor EAX, EAX
.LBB2_4: ; %bb
mov ECX, DWORD PTR [ESP + 32]
cvttsd2si ECX, QWORD PTR [ECX + EAX]
mov ESI, DWORD PTR [ESP]
lea ESI, DWORD PTR [ESI + EAX]
test DL, 1
je .LBB2_15
.LBB2_5: ; %bb1.preheader
mov EBX, DWORD PTR [ESP + 8]
mov EDI, ECX
.LBB2_6: ; %bb1
movsd XMM0, QWORD PTR [ESI]
cvtsi2sd XMM1, ECX
ucomisd XMM1, XMM0
jbe .LBB2_8
.LBB2_7: ; %bb2
cvttsd2si ECX, XMM0
.LBB2_8: ; %bb3
cvtsi2sd XMM1, EDI
ucomisd XMM0, XMM1
jbe .LBB2_11
.LBB2_9: ; %bb3.cBB7_crit_edge
mov EBP, 4294967295
.align 16
.LBB2_10: ; %cBB7
inc EBP
je .LBB2_16
.LBB2_11: ; %bb5
add ESI, 80
dec EBX
jne .LBB2_6
.LBB2_12: ; %bb7
sub EDI, ECX
cvtsi2sd XMM0, EDI
mov ECX, DWORD PTR [ESP + 36]
movsd QWORD PTR [EAX + ECX], XMM0
add EAX, 8
dec DWORD PTR [ESP + 4]
jne .LBB2_4
.LBB2_13: ; %return
add ESP, 12
pop ESI
pop EDI
pop EBX
pop EBP
ret
.LBB2_14: ; %bb.nph16
lea EDX, DWORD PTR [EAX - 1]
mov DWORD PTR [ESP + 8], EDX
cmp EAX, 1
setg DL
jmp .LBB2_2
.LBB2_15: ; %bb.bb7_crit_edge
mov EDI, ECX
jmp .LBB2_12
.LBB2_16: ; %bb4
cvttsd2si EDI, XMM0
jmp .LBB2_10
_Process endp